模型鲁棒性的奇异值视角

Dec, 2020

A Singular Value Perspective on Model Robustness

Malhar Jere, Maghav Kumar, Farinaz Koushanfar

TL;DR本文研究探讨卷积神经网络在人工智能领域的应用，并且探讨该类网络存在的问题和局限性，提出使用奇异值分解方法可以分析卷积神经网络中涉及到的非人类因素，从而对其产生作用的原因展开研究。同时提出了一种新的特征分配方法，建立了卷积神经网络的理论框架。

Abstract

convolutional neural networks (CNNs) have made significant progress on several computer vision benchmarks, but are fraught with numerous non-human biases such as vulnerability to adversarial samples. Their lack o